Upload
others
View
10
Download
0
Embed Size (px)
Citation preview
1© 2015 The MathWorks, Inc.
The Rise of Engineering-Driven
Analytics
Roy Lurie, Ph.D.
Vice President Engineering,
MATLAB Products
2
The Rise of Engineering-Driven Analytics
3
The Rise of Engineering-Driven Analytics
4
Big DataCompute
Power
Machine
Learning
Limited
users, scope
& technology
Pervasive
users, scope,
& technology
• Engineering
• Business
• Transactional
• Desktop -Multicore, GPU
• Clusters
• Cloud computing
• Hadoop with Spark
• Neural Networks
• Classification
• Clustering
• Regression• …and much more…
Analytics are now pervasive
Apply robust, statistically-motivated methods to data produced from complex systems to understand what has happened,
predict what will happen, and
suggest decisions or actions.
Descriptive & Diagnostic
Predictive
Prescriptive
5
Business Data
Social profile
Geolocation
Keystroke logs
Transactions
Engineering DataImages
Predictive Model
Offer to Customer
IMPROVED
Use Image Processing
to add image data to the model,
improving performance
Analytics in e-commerce
6
Transactions
Keystroke logs
Geolocation
Social profile
Sensor
Images
Audio
Video
Using now
Source: Gartner Big Data Industry Insights, March 2016
Engineering Data
Business Data
Level of Industry / User Adoption
Consider the Data in Data Analytics
7
Transactions
Keystroke logs
Geolocation
Social profile
Sensor
Images
Audio
Video
Using now
Planned
Source: Gartner Big Data Industry Insights, March 2016
Engineering Data
Business Data
Level of Industry / User Adoption
Consider the Data in Data Analytics
8
The Rise of Engineering-Driven Analytics
9
Architecture of an analytics system
Data from businesssystems
Data from instruments and connected systems
Analyticsand MachineLearning
10
Architecture of an analytics system
Data from businesssystems
Data from instruments and connected systems
Analyticsand MachineLearning
Predictive Model deployed in smart and embedded systems
Predictive Model deployed on cloud and business systems
MATLAB & Simulink Integrates in Embedded Systems and Enterprise
IT Workflows
11
12
25% cost reduction
13
Example – BuildingIQAdaptive building energy management
14
Optimizing Energy Costs and Consumption at Building IQ
Predictive Model deployed on cloud with client system
and real-time data feeds
DATA - Billions of data points:Physics, energy cost, power, internal temperatures,
ambient temperatures, ambient humidity, building
operation schedule, comfort bounds, etc.
Weather
Feeds
Current energy
costs & demand
Analytics and
Machine Learning
plus system identification,
control theory & more
MATLAB Toolboxes Just Work –
and work together!
15
We could rapidly translate our prototypes into production algorithms that deal reliably with real-world noise and uncertainty
Borislav Savkovic, BuildingIQ
Why MATLAB?
Robust numerical algorithms
Extensive visualization and analytics tools
Industry-robust and reliable mathematical optimization routines
Good object-oriented framework
Ability to interface with Java (for backend work)
Running MATLAB in the cloud in production
Unit-testing framework
MATLAB Impeccable Numerics for Trusted Results
Why MATLAB?
Robust numerical algorithms
Industry-robust and reliable mathematical optimization routines
16
Example – ScaniaAutomatic emergency braking using sensor fusion and analytics
17
18
Using Model-Based Designto build and deploy the analyticsin an embedded control system
MATLAB Integrates Analytics and
Model-Based Design
19
Implementing Sensor Fusion at Scania
Predictive Model deployed on vehicle
Vehicle logs
of video and radar data
Machine learning
to develop fusion algorithms
for situation detection
20
The Rise of Engineering-Driven Analytics
Medical Devices
AeronauticsOff-highway vehiclesAutomotive
Oil & GasIndustrial Automation Clean Energy
Retail Finance Healthcare management Internet
21
Sensor Data (~1 minute)10-100 sensors/machine
Quality State (~40 minutes)
Classification using Statistics, Machine Learning, and Neural Networks
Predictive Maintenance for polymer-based production machines
22
Deployment – a MATLAB App used by machine operators
M153
M157
State OK
State NOT OK
23
24
The need for data scientists
Domain expertise
Coding and integration skills
Statistical and mathematical
knowledge
25
What they say• Expand university programs• Train existing analysts
26
27
TSMC Student Contestuse process control data to improve semiconductor yields
21 teams competed
Wafer Big Data in Hadoop
MATLAB used by winning team and 2nd place team
28
Built-in MATLAB analysis
Simulink support via Raspberry Pi
IoT open data platform for students and makers
29
MATLAB lets you be your own data scientist
MATLAB is Designed and
Documented to be Easy for
Engineers and Scientists to UseDomain
expertise
Statistical and
mathematical
knowledge
Coding and
integration skills
30
Big DataCompute
Power
Machine
Learning
Limited
users, scope,
& technology
Pervasive
users, scope,
& technology
• Engineering
• Business
• Transactional
• Neural Networks
• Classification
• Clustering
• Regression• …and much more…
In MATLAB
NEW for MATLABAudio System Toolbox R2016aVision HDL Toolbox R2015a
• Native support for engineering data
• Database interfaces
• Streaming
• DesktopMulticore, GPU
• Clusters
• Cloud computing
• Hadoop with Spark
31
Big DataCompute
Power
Machine
Learning
Limited
users, scope,
& technology
• Native support for engineering data
• Database interfaces
• Streaming
• Datastoretext, image, video, Excel files
• Timetable, string, and tall arrays 2016b
Pervasive
users, scope,
& technology
• Engineering
• Business
• Transactional
• DesktopMulticore, GPU
• Clusters
• Cloud computing
• Hadoop with Spark
• Neural Networks
• Classification
• Clustering
• Regression
In MATLAB
32
• Native support for engineering data
• Database interfaces
• Streaming
• Datastoretext, image, video, Excel files
• Timetable, string, and tall arrays 2016b
Big DataCompute
Power
Machine
Learning
Limited
users, scope,
& technology
Pervasive
users, scope,
& technology
• Engineering
• Business
• Transactional
• Neural Networks
• Classification
• Clustering
• Regression
In MATLAB• Multicore & GPU
• MATLAB Distributed Computing Server and EC2 Support
• Hadoop with Spark support R2016b
• MATLAB Production Server
MATLAB is fast:
- heavily optimized libraries
- JIT compiled
- takes advantage of the
compute power you have
• DesktopMulticore, GPU
• Clusters
• Cloud computing
• Hadoop with Spark
33
Big DataCompute
Power
Machine
Learning
Limited
users, scope,
& technology
Pervasive
users, scope,
& technology
• Engineering
• Business
• Transactional
• Neural Networks
• Classification
• Clustering
• Regression
In MATLAB • Statistics and Machine Learning Toolbox
• Classification Learner App R2015a
• Neural Network Toolbox
• CNNs for Deep learning R2016a
• Machine learning with code generation
• Multicore & GPU
• MATLAB Distributed Computing Server and EC2 Support
• Hadoop with Spark support R2016b
• MATLAB Production Server
• Native support for engineering data
• Database interfaces
• Streaming
• Datastoretext, image, video, Excel files
• Timetable, string, and tall arrays 2016b
• DesktopMulticore, GPU
• Clusters
• Cloud computing
• Hadoop with Spark
34
Classification Learner App in Statistics and Machine Learning Toolbox
35
Deep Learning with Neural Network Toolbox - New in R2016a
camera = webcam;
img = snapshot(camera);
net = alexnet;
label = classify(net,img)
Technical Computing and Data Analytics
36
Example –First consumer otoscope in a mobile device using machine learning and computer vision
37
The Rise of Engineering-Driven Analytics
Be your own Data Scientist!
Big DataCompute
PowerMachine Learning
Limited users, scope, & technology
Pervasive users, scope, & technology